OPEN MODEL
by Meta
Meta's family of open-weight large language models, released in sizes from small models that run on a laptop to large multimodal models for servers. Businesses download the weights to self-host, fine-tune for their own tasks, or use managed versions on major cloud platforms.
- PRICE
- Free
- SET-UP
- Advanced
- DATA
- Can run onshore
- FOUNDED
- 2023
- HQ
- Menlo Park, USA
- WEBSITE
- llama.com
- REVIEWED
- October 2026
SPEC SHEET
001 / WHAT IT DOES
- 01Downloadable weights in sizes ranging from edge devices to large servers
- 02Multimodal models that accept images as well as text
- 03Fine-tuning support through common open-source training tools
- 04Managed hosting through major cloud platforms and inference providers
- 05Community licence permitting commercial use with conditions
BEST FOR
- Organisations wanting a private chatbot on their own infrastructure
- Developers fine-tuning models for specialist vocabulary
- Teams deploying small models on devices or on-premises servers
TYPICAL USES
002 / THE HONEST TAKE
STRENGTHS
- +Free commercial use for the vast majority of businesses
- +Wide range of sizes, from edge devices to large servers
- +Large ecosystem of fine-tunes, tools and hosting options
- +Available as a managed model on major clouds
WATCH OUTS
- –Licence is custom and more restrictive than true open source
- –Self-hosting larger models requires GPU expertise and budget
- –Newer frontier capability from Meta is not released as open weights
003 / FOR AUSTRALIAN BUSINESSES
Not by default, but it can be kept in Australia by self-hosting or using it through an Australian cloud region.
Llama weights run wherever you deploy them, so data residency is decided by your hosting choice rather than by Meta. Self-hosting on your own servers or Australian cloud GPU instances keeps prompts onshore and away from Meta. Managed Llama services vary by provider and region, so check that the specific model you want is offered in an Australian region before assuming local processing.
PRIVACY ACT & PRACTICAL NOTES
Self-hosting Llama on your own servers or GPU instances in an Australian cloud region keeps prompts and outputs onshore, which helps with Privacy Act 1988 obligations and client data requirements. Check regions before choosing a managed option, as AWS Bedrock serves its Llama models from US regions. Read the licence rather than assuming it is fully open source, because it includes an acceptable use policy and naming conditions. Meta's newest frontier model has been released as a closed service, so check whether the open Llama line still meets your capability needs before committing.
004 / WHAT IT COSTS
Free
Weights are free to download under Meta's Llama community licence, which allows commercial use but has conditions, including a cap for very large platforms and attribution rules. You pay for your own compute, or per token when using a hosted version through a cloud provider. See llama.com.
Check current prices on llama.com ↗︎005 / CONNECTS WITH
Want it wired into your CRM, inbox or Xero? That's what we build.
006 / ALTERNATIVES
Also worth a look
- QwenOnshore possibleFreemium→
- DeepSeekOnshore possibleFreemium→
- Hugging FaceOnshore possibleFreemium→
- Google AI StudioOnshore possibleFreemium→
- OpenRouterOffshoreFreemium→
- OllamaOn deviceFreemium→
- LangChainAU optionFreemium→
- CohereOnshore possibleFreemium→
Also compared against: Mistral, Google Gemma.
007 / QUESTIONS
Anything else is quicker to answer on a call. Thirty minutes, free, with the people who would build it.
For most businesses, yes. The Llama community licence permits commercial use, but it includes an acceptable use policy, attribution requirements and a separate licence requirement for very large platforms. Have someone read the licence for the specific release you deploy, because terms can differ between releases.
We set up and connect tools like this for Australian businesses, with the privacy settings done right from day one.
Book a free call→